

531 episodes
by Enoch H. Kang · English
Cut through the noise. We curate and break down the most important AI papers so you don’t have to.

Pluralistic AI alignment using Direct Preference Optimization when user preferences vary in unobserved ways; argues standard binary preference data (A vs B) is mathematically insufficient and…
24 Oct 2025 · 12m · 9 chapters
The episode explains why next-token pre-training that minimizes cross-entropy (or sequence-level KL) can still fail after fine-tuning for coding/reasoning, and introduces the “coverage principle.” It…
24 Oct 2025 · 16m · 12 chapters
The episode argues alignment and personalization for AI assistants should shift from static RLHF (human-annotated A/B comparisons) to RLHI, reinforcement learning from human interaction using real…
24 Oct 2025 · 14m · 6 chapters
Early experience for training autonomous language agents—using the agent’s own explorations and failures to learn without dense external rewards, bridging imitation learning (IL) and reinforcement…
24 Oct 2025 · 13m · 5 chapters
Demystifies emergent exploration in goal-conditioned RL via Single Goal Contrastive RL (SGCRL), arguing exploration comes from learned representation structure (not bigger networks), using InfoNCE…
22 Oct 2025 · 15m · 11 chapters
“Rewriting history” interventional analyses for causal study of how pretraining data affects LLM factual knowledge.
22 Oct 2025 · 19m · 7 chapters
Defines AGI as an AI matching/exceeding a well-educated adult’s cognitive versatility and proficiency, then proposes a quantifiable “standardized AGI score” (0–100%) using CHC (Cattell–Horn–Carroll)…
22 Oct 2025 · 16m · 12 chapters
The episode explains “learning from language feedback” (LLF) and a new theory proving it can be exponentially faster than learning from scalar rewards.
21 Oct 2025 · 20m · 11 chapters
Efficient “pure exploration” when ground truth is costly—choosing the fewest adaptive queries/tests to identify the correct hypothesis with high confidence or within a fixed budget.
21 Oct 2025 · 17m · 8 chapters
How preference variance (P-VAR) can be used to select higher-value preference data for DPO/LLM alignment, improving efficiency and final model quality.
20 Oct 2025 · 15m · 8 chapters
The episode argues that today’s LLM coding assistants fail when they generate overly long, low-confidence code, and presents “Empower” (Logit Threshold Empowerment) as a scalable way to train agents…
20 Oct 2025 · 14m · 6 chapters
Richard Sutton argues LLMs are a “dead end” and will be replaced by continual-learning, goal-driven agents built from reinforcement learning (RL), not text imitation.
20 Oct 2025 · 13m · 3 chapters
How reinforcement learning (RL) can train LLM-based agents for multi-step, tool-using problem solving, and why training recipe choices beat raw model scale.
19 Oct 2025 · 15m · 11 chapters
How to tell when multi-agent LLM “synergy” is truly emergent (goal-directed coordination) versus just bots oscillating, and how prompt design can steer it.
19 Oct 2025 · 14m · 5 chapters
Learning-to-measure (L2M) for active feature acquisition, aiming to learn a general policy that decides which next data feature to measure to reduce prediction error while accounting for real-world…
19 Oct 2025 · 16m · 10 chapters
Andrej Karpathy argues AGI is likely about a decade away and that progress depends on missing “agent” capabilities (continual learning, robust multimodality, and generalized computer use), plus…
19 Oct 2025 · 16m · 12 chapters
When to teach LLM reasoning—whether to “front-load” reasoning data during pretraining or bolt it on later via supervised fine-tuning (SFT).
18 Oct 2025 · 13m · 6 chapters
Representation-based exploration (REPEX) for language models, combining RL with a novelty bonus to overcome the “sharpening constraint” and “diversity collapse,” improving discovery beyond just…
18 Oct 2025 · 17m · 11 chapters
LLM jailbreaking and prompt-injection defenses are evaluated incorrectly; adaptive attackers bypass “state-of-the-art” defenses with >90% attack success rates, creating a false sense of security.
18 Oct 2025 · 16m · 11 chapters
In-context learning (ICL) generalizing out of task distribution depends on “task diversity,” not just the number of tasks.
16 Oct 2025 · 20m · 8 chapters
How to scale the reinforcement learning (RL) stage of LLM training with a principled compute-scaling framework (“Scalar RL”), replacing guesswork with predictable A-B scaling curves.
16 Oct 2025 · 14m · 5 chapters
A joint study (Anthropic, UK AI Security Institute, Alan Turing Institute) shows LLM “data dilution” defenses fail: a fixed small number of poisoned documents can install a backdoor regardless of…
16 Oct 2025 · 14m · 4 chapters
Dual Goal Representations (DGR) for goal-conditioned reinforcement learning (GCRL).
14 Oct 2025 · 17m · 9 chapters
The episode argues AI is shifting from learning from human data (LLMs trained on online text) toward an “era of experience,” where agents generate their own training data through long-term…
14 Oct 2025 · 17m · 9 chapters